What are the implications of AI-powered bookkeeping like Zeni and Digits for the cost and accuracy of bookkeeping, and the customer experience?

I haven't thought a whole lot about it. When I was at Pilot, the feeling was very much that we need to automate things, but we don't necessarily need machine learning to do that. It's just literally turn it into a script that you can run that's entirely deterministic. However, there were a few things that I got a real appreciation for working at Pilot. 

The first is that bookkeeping is really hard. When I started doing it, I was assuming that it's not really hard, and not really a skill thing. I learned, no, these people are actually really smart. 

The other thing I learned was, you really have to be correct. 

Just like how a surgeon or a lawyer needs to be doing things the right way, a bookkeeper also needs to do the same. So the threshold that your AI has to cross is very high. The bar is set very high there. It's not like content marketing—that would be very high on my list of jobs that the robots are coming for, because you can literally just plagiarize content that's out there on the internet and give it to someone and call that a product.

Bookkeeping is a lot harder than that to solve with AI, because you need to be able to explain to a customer why this is the right category, why we ticked and tied this transaction to the right other transaction, and the answer to that cannot be, "Because the algorithm said." 

The answer has to be based on bookkeeping best practices, which someday, you may be able to do an AI that really understands bookkeeping. When you ask that question, this is the answer you get. 

But there will be a lot of skepticism there. Just like if you asked your lawyer a question, you're expecting to get an answer that's grounded in the law. So I'm naturally not super bullish on AI for a lot of things. Bookkeeping is one of the things that I don't think is super close.